The Intelligence Extracting the HS Code from a Complex Product Description in Seconds
In the world of customs clearance, every product has an HS code (GTİP), and correctly identifying this code determines the difference between an error-free declaration and a serious financial risk; but extracting the correct code from complex product descriptions, invoices, and declarations is an expert, slow, and error-prone job. For Solmaz Lojistik, a well-established player in international transport and customs clearance, we restructured the Genius platform end-to-end to automate this task. We transformed it into a structure that extracts the correct HS code from complex product descriptions in seconds, finds the correct result even in faulty searches, and centralizes scattered commercial data.
Featured Results
As Phexum, we conducted all business analysis from the very beginning of the project, completely redesigned and developed the platform's software architecture and core algorithms. We built existing operational flows from scratch with much faster and more sustainable technologies, taking the user experience to the highest level with modern interfaces. With our NLP models, we ensured the correct HS code is extracted accurately from complex product descriptions in seconds; with fuzzy logic, we made it possible to list the most accurate result even in incorrectly or incompletely typed searches; we established an infrastructure facilitating the transfer of AI models across environments, and we moved scattered commercial data to a centralized architecture that collects, deduplicates, and processes it.
Who the Client Is and the World They Live In
Solmaz Lojistik is one of the industry's well-established players in international transport, customs clearance, and supply chain management. The fundamental difficulty of this world is the sensitive and highly labor-intensive nature of customs clearance. Every product must be declared with the correct HS code; a wrong code leads to delays and financial risks. However, determining this code often requires interpreting non-standard product descriptions in free text, and doing this manually is both slow and dependent on an expert. On top of this comes data scatter: non-standard commercial data flowing from different channels must be collected, cleared of repetitions, and made processable. This is one of the most labor-intensive processes of logistics operations. This was exactly the problem Genius was reborn to solve.
Cross-Cutting Concepts and Architecture
The first concept running through this entire platform is going from free text to the correct code. A product description is a scattered, non-standard text written by a human; the system's job is to reach the single correct HS code from this text. The technical equivalent of this is natural language processing, and specifically keyword extraction: extracting the meaningful parts that determine the correct classification from a complex description. Our NLP models do this autonomously from invoices and declarations and match it with the system accurately in seconds. Right next to this stands the search experience; because users do not always type the exact correct term. Here fuzzy logic steps in: finding the most accurate logistics and customs data even with a faulty or incompletely typed term. The benefit is clear: a slow, expert-dependent detection job turns into a flow where everyone reaches the correct result in seconds.
The second concept is making the data and the model manageable. An AI system shouldn't just predict well; it must also be continuously updatable. Therefore, we established an infrastructure that allows models predicting HS codes to be transferred losslessly, securely, and quickly across different server environments, namely development, staging, and production; we developed custom interfaces where new models can be easily created and tested. This is called the MLOps approach, which manages the lifecycle of models. Beneath this lies the data layer: we accelerated all the data collection, HS code-based deduplication, simplification, and processing processes of raw commercial data—which slow down field operations—with high-performance data pipelines and gathered them in a single center. Ultimately, both the AI models and the data feeding them became continuously and sustainably manageable.
How the Collaboration Began
For Solmaz Lojistik, Genius was an application of strategic importance, and we restructured it end-to-end. Our role was not merely to patch an existing system, but to fundamentally rebuild the platform: we conducted detailed business analysis from the very beginning of the project, redesigned the architecture and core algorithms from scratch. This was a job that required deeply understanding the reality of the field; because in a rule-based and sensitive area like customs clearance, the right solution could only be built by grasping the domain well.
Building Blocks of the Solution
Smart HS code matching autonomously extracts the correct code from complex product descriptions using NLP and matches it accurately in seconds. The smart search and fuzzy logic layer perfects the search experience by listing the most accurate customs data even for faulty or missing terms. The modular AI and environment portability infrastructure ensure safe transfer of models across environments and easy creation of new models. The central data architecture makes the entire process manageable from a single point with high-performance pipelines that collect, deduplicate, and process scattered commercial data. All of these rise on a modern architecture and modern interfaces rebuilt from scratch.
No longer a task dependent on an expert, but an automatic flow lasting seconds.
The Impact We Created
Reborn with Phexum's architecture, R&D, and analytical power, Genius multiplied Solmaz Lojistik's customs clearance and operational speed, while minimizing time losses and errors stemming from manual HS code detection and data management. The path from complex product description to the correct code is no longer a task dependent on an expert, but an automatic flow lasting seconds. Placing AI and NLP right at the center of logistics processes, this project became one of the most concrete examples of data-driven corporate transformation in the sector.